Repository files navigation

In this repository you can see 2 main programs:car_counter_yolov3_custom_classes.py andcar_counter_yolov3_COCO_6_classes.py

The first one is a lighter version of the second. Basically, I`ve trained YOLOv3 to detect 5 classes:

• sedan
• minivan
• SUV
• hatchback
• universal

But, to be honest,.weights file that I got in the end is pretty wack and works not that good on different videos. But it's still here.

• Downloadyolo-obj_final.weights file for YOLOhere
• Download any test-video with cars driving around and put it tovideos/ folder (or use any of those that are already there)
• Move.weights file toyolo/ folder
• Go to the project's repository via command line
• typepython car_counter_yolov3_custom_classes.py -y yolo --input videos/THE_NAME_OF_YOUR_TEST_VIDEO --output output --skip-frames 5 and hitEnter

The proccessed video will be saved to theoutput/ folder

The second one uses pretrained.weights file fromthis site. So I didn't need to train YOLOv3 myself once again. This program can:

• detect and track objects of all of 80 COCO classes
• count objects of each of 6 classes:
  • car
  • truck
  • person
  • motorcycle
  • bicycle
  • bus
• count the amount of all of those objects on each frame of the video
• put the results into.json file

• DownloadYOLOv3-608.weights file for YOLOhere

• Download any test-video with cars driving around and put it tovideos/ folder (or use any of those that are already there)

• Move.weights file toyolo/ folder

• Go to the project's repository via command line

• typepython car_counter_yolov3_COCO_6_classes.py -y yolo --input videos/THE_NAME_OF_YOUR_TEST_VIDEO --output output --skip-frames 5 and hitEnter

  You can change theskip-frames parameter (the higher it is, the faster the program works). But the accuracy will be lower

  The proccessed video and the.json file will be saved to theoutput/ folder